Microsoft launches Copilot Cowork for task execution
Microsoft has made Copilot Cowork available through its Frontier programme. The tool is designed for long-running, multi-step work in Microsoft 365.
The launch expands Microsoft's push to add more agent-style AI functions to its workplace software, targeting tasks that span files, tools and routine business processes.
Copilot Cowork lets users describe a desired outcome, then generates a plan to complete the work. The system tracks progress and lets users intervene during the process, rather than only returning a final answer.
The product includes skills from Claude and Microsoft, including calendar management and daily briefings. It can be used for single tasks as well as recurring work such as a monthly budget review.
Early access
Capital Group was among the organisations given early access to the product. According to Microsoft, the investment management firm used it for planning, scheduling, creating deliverables and preparing for executive reviews.
"We have been using Copilot since its launch in 2024, and the new capabilities in Cowork will help us automate and scale the Copilot ecosystem. This isn't about generating content or answers. It's about taking real action-connecting steps, coordinating tasks, and following through across everyday workflows. Because Cowork operates on our enterprise data and within our security and risk boundaries, we can experiment, learn, and scale with confidence. That allows us to move faster and focus AI in places where it actually delivers value," said Barton Warner, Senior Vice President of Enterprise Technology at Capital Group.
The rollout also includes updates to Researcher, another Microsoft 365 Copilot feature focused on complex information gathering and analysis. Researcher synthesises information from multiple sources and produces cited responses for business use.
Model review
A new Critique function in Researcher uses more than one model in sequence. One model plans the task and drafts a response, while another reviews and refines the output before the final report is produced.
This approach draws on models from Frontier labs including Anthropic and OpenAI. Microsoft said the change lifted Researcher's score by 13.8% on the Deep Research Accuracy, Completeness, and Objectivity benchmark, which it described as an industry standard for deep research quality.
Researcher is also gaining a feature called Model Council. It lets users compare answers from different models side by side to identify points of agreement, disagreement and unique contributions from each system.
The announcements reflect Microsoft's broader strategy of building Microsoft 365 Copilot around access to multiple model providers rather than relying on a single model family. The company has positioned that approach as a way to bring different forms of AI reasoning into workplace software already used for communication, scheduling, writing and analysis.
Copilot Cowork follows Microsoft's earlier disclosure that it would bring the technology platform behind Claude Cowork into Microsoft 365 Copilot. With this move, that work is now available to customers in the Frontier programme, which serves as an early-access route for new Microsoft AI products.
The update comes as large technology companies compete to make AI systems more useful inside office software by moving beyond chat responses into task execution. Rather than simply generating text, the focus has shifted to systems that can handle sequences of actions, interact with business data and support repeatable workflows under existing security controls.
Microsoft said Copilot Cowork operates on enterprise data and within organisational security and risk boundaries.